Guide
Scoring & weighting
How the 12 pillar scores roll up into a single Appkitekt AI Index — and why the weighting shifts by brand, category and asset kind. We explain the logic; the exact weights stay inside the product.
In plain English
Each of the 12 pillars gets its own score out of 100. We then combine them into a single score out of 100 for the whole brand — the Appkitekt AI Index. Because a hotel booking site, a mobile game, and a payments API are winning on very different things, we count some pillars more than others depending on what you actually are. The math is consistent; the emphasis shifts.
The roll-up, technical version
We catalog and score the signals that AI agents actually use to discover, select, and trust brands — signal selection is observed from agent behavior. Every pillar produces a 0–100 readiness score, and the Appkitekt AI Index is a single 0–100 number that combines them all. Each pillar carries a base weight, adjusted by where you are (category) and how confident the underlying sample is. The structure of the roll-up is identical across web, app and MCP audits — the per-pillar emphasis is not.
Two surfaces, one frame
Web, app and MCP audits all land on a single 0–100 score with the same band labels (Invisible → Default-of-record) and the same confidence treatment. Two things differ between them:
- The factor floor — 130+ factors on a web audit, 200+ on an app audit, 210+ on an MCP / API audit, each with one mapped factor per agent in the panel. This is a coverage count, not a score input.
- The relative emphasis per pillar — AAIO, AXO, PULSE and RIVAL carry more weight in app mode; AEO, GEO and SIGNAL carry more weight on the web; MCP, AAIO and AXO carry the most on MCP audits.
Relative emphasis per surface
We don't publish the exact percentage each pillar contributes — that's tuned per category, per surface and per agent generation, and it shifts between cycles. Instead, the table below shows the relative emphasis each pillar carries on a four-step scale (Heavy, Medium, Light, Minimal). It's enough to understand why the same audit can score differently on web, app and MCP surfaces.
| Pillar | Web | App | MCP / API | Why it shifts |
|---|---|---|---|---|
| AEO | Heavy | Medium | Minimal | Web rewards answer-shape pages; app rewards listing structure; barely used in MCP. |
| GEO | Medium | Light | Light | App listings are short; MCP servers are described, not summarized. |
| SIGNAL | Medium | Light | Light | Web co-citations dominate; app press/lists secondary; MCP registries weigh in. |
| PULSE | Light | Heavy | Light | Version cadence and what's-new matter most in app; MCP tracks schema-change cadence. |
| RIVAL | Medium | Heavy | Medium | Rival-app overrides hit installs; in MCP, tool-name lookalikes steal calls. |
| AAIO | Light | Heavy | Heavy | App Intents in app mode; tool-action surface in MCP mode. |
| AXO | Medium | Heavy | Heavy | AASA / Universal Links handoff in app; in MCP, predictable errors and auth scopes. |
| PROTO | Light | Light | Medium | Well-known / spec tracking is heaviest on MCP audits. |
| BUG | Light | Light | Medium | MCP adds tool-description hallucination risk on top of the standard grade. |
| DAO | Light | Minimal | Light | Token efficiency matters everywhere; MCP responses are pure token cost. |
| MCP | Light | Light | Heavy | Native surface for MCP audits — manifest, tool schemas, auth scopes, registry presence. |
| SOCIAL | Light | Light | Minimal | Grounding signal across all surfaces; lightest on pure tool servers. |
Tiers are directional. A pillar tagged Heavy contributes meaningfully more to the Index than one tagged Light; two pillars in the same tier are not guaranteed to carry the same numeric weight.
Category modifiers
The weighting is tuned per category because the failure mode that hurts you most varies by what you do. Below are directional examples — not the actual tuning tables.
Web categories:
- Consumer brands & content sites — discovery pillars carry more weight; AAIO/AXO carry less.
- SaaS with public APIs / MCP — AAIO and AXO carry more weight; PULSE carries less.
- Marketplaces & comparison-driven categories — RIVAL and DAO carry more weight.
- News & high-volatility data — SIGNAL and BUG carry more weight.
App categories:
- iOS / Android games — RIVAL and PULSE carry more weight.
- Productivity & utility apps — AAIO and AXO carry more weight (App Intents critical).
- Finance, health, kids apps — AXO and BUG carry more weight (trust + hallucination risk).
- Subscription consumer apps — PULSE and RIVAL carry more weight (ratings velocity, rival churn).
MCP / API vendors:
- Pure tool servers (data, search, automation) — MCP, AAIO and AXO dominate (manifest, tool schemas, auth scopes, error shape).
- SaaS exposing MCP alongside a UI — RIVAL and PROTO carry more weight (lookalike tool names, fast-moving spec).
- Regulated APIs (finance, health, identity) — AXO, BUG and SIGNAL carry more weight (auth, hallucinated capabilities, registry trust).
Confidence
A pillar with a small or partial sample contributes proportionally less to the Index. This prevents a single rate-limited model from skewing the headline number. The dot indicators on each pillar tile show the confidence state used in the current roll-up.
Why your Index can move without any push
Agent behaviour shifts between cycles even when you change nothing — models get updated, indexes get refreshed, rivals push their own fixes. Single-cycle moves of ±3 points on the Index are noise. Treat ±7 or two-cycle trends as real, and open the per-pillar view to attribute the move. In app mode the Index can also move on store-side editorial pushes (App Store featuring, category re-ranks) outside our control.
From score to dollars
The Index is a readiness number — it doesn't claim a revenue figure on its own. To project agentic traffic, human impressions and a USD revenue-at-stake band, the Index and per-agent breakdown feed the reasoning in Traffic & revenue. Every output is shown as a low–high band.